The goals of this project are to elucidate the fundamental principles underlying protein biochemical function and to apply the resulting insights on the interrelationships of protein structure, function and evolution to functional annotation and to assist in drug discovery. The underlying theme is that many features of protein structure and function arise from their physical properties without selection for function, which evolution then acts to fine tune/optimize. This will be demonstrated by the coincidence of the structural and functional properties of native proteins with artificially generated, homopolypeptide SYN protein structures to which protein-like sequences are added based on their stability in a particular fold. We will focus on ligand binding pockets and will explore the relationship between global fold similarity, pocket location/shape and amino acid conservation. A key objective is to demonstrate that ligand binding pocket geometry and sequence can be uncoupled from a protein's global structure, so that functional inference can be made between structurally different proteins. These ideas will extend our FINDSITEcomb Ligand Homology Modeling algorithm by developing approaches that better identify common ligand binding pockets and ligands in proteins having either similar or unrelated global folds. FINDSITEcomb will provide predicted structures, GO function, ligand binding sites, and virtual ligand screening/binding pose predictions. An important application will be to predict off-targets of FDA approved drugs in the human proteome. These off-target predictions will be experimentally tested for a significant number of proteins. To achieve these objectives, four Specific Aims are proposed: 1. Examination of the roles of physics and evolution in determining protein structure and function. 2. Exploiting insights from Aim 1, a new approach to difficult target threading will be developed. 3. The major limitations of FINDSITEcomb will be addressed and the methodology applied to drug repurposing. 4. Using thermal shift assays, FINDSITEcomb's virtual screening predictions will be experimentally tested. The entire computational methodology will be applied to the human proteome and model organisms, and the SUNPRO database will report all results. All tools will be made available on our superPSIFR webserver and as downloadable software, including source code. These aims represent an integrated effort to elucidate the principles underlying protein structure and function and to apply these insights to improve function predictions.